| Abstract: | Cloud cover information is used alongside weather forecasts in various fields of research; however, ground observation of cloud cover is conducted by human observers, a method that is subjective and has low temporal and spatial resolutions. To address these problems, we have developed an improved algorithm to calculate cloud cover using sky image data obtained with Skyviewer equipment. The algorithm uses a variable threshold for the Red Blue Ratio (RBR), determined from the frequency distribution of the Green Blue Ratio (GBR), to calculate cloud cover more accurately than existing algorithms. To verify the accuracy of the algorithm, we conducted daily, monthly, seasonal, and yearly statistical analyses of human observations of cloud cover, obtained every hour from 0800 to 1700 Local Standard Time (LST) for the entirety of 2012 at the Gangwon Regional Meteorological Administration (GRMA), Korea. A case study compared daily images taken at 1200 LST in each season with pixel images of cloud cover calculated by our improved algorithm. The selected cases yielded a high correlation coefficient of 0.93 with the GRMA data. A monthly case study showed low root mean square errors (RMSEs) and high correlation coefficients (Rs) for December (RMSE = 1.64 tenths and R = 0.92) and August (RMSE = 1.43 tenths and R = 0.91). In addition, seasonal cases yielded a high correlation of 0.9 and 87% consistency within ± 2 tenths for winter and a correlation of 0.83 and 82% consistency for summer, when cases of cloud-free or overcast conditions are frequent. Annual analyses showed that the bias of GRMA and Skyviewer cloud cover data for 2012 was -0.36 tenth, and the RMSE was 2.12 tenths, with the GRMA data showing more cloud cover. Considering that the GRMA and Skyviewer data were gathered at different spatial locations, GRMA and Skyviewer data were well correlated (R = 0.87) and showed a consistency of 80% in their cloud cover data (consistent within ± 2 tenths). |
